[stingraycharles]

No, it's just that it's confusing, because there are two ways of using Claude Code credentials:

1. Take the oauth credentials and roll your own agent -- this is NOT allowed

2. Run your agentic application directly in Claude Code -- this IS allowed

When OpenClaw says "Open-Claw style CLI usage", it means literally running OpenClaw in an official Claude Code session. Anthropic has no problems with this, this is compliant with their ToS.

When you use Claude Code's oauth credentials outside of the claude code cli Anthropic will charge you extra usage (API pricing) within your existing subscription.

[filleokus]

But... Even when running it in mode 2 ("claude -p") they at certain points tried to detect OpenClaw-usage based prompts made, and blocked them [0]. Now OpenClaw says that Antrophic sanctions this as allowable again.

I agree with GP that this is hard to take seriously.

[0]: https://x.com/steipete/status/2040811558427648357

[Wowfunhappy]

> Even when running it in mode 2 ("claude -p") they at certain points tried to detect OpenClaw-usage based prompts made, and blocked them [0].

But then the Claude Code product manager said:

> This is not intentional, likely an overactive abuse classifier. Looking, and working on clarifying the policy going forward.

https://xcancel.com/bcherny/status/2041035127430754686#m

[zaphar]

I mean, if you are them and trying to detect when people are using your system incorrectly the detection system is going to be a little bit flaky. How do they prove you aren't violating your ToS by using OAuth for a system they didn't approve that usage for?

The fault here is not with Anthropic. It lies with cowboy coders creating a system that violates a providers terms of service and creating an adverse relationship.

[stingraycharles]

I have never heard of this, and cannot be reproduced, and is not according to Anthropic's ToS. And there's a lot of FUD being spread around.

They don't ban Openclaw prompts, each custom LLM application provides a client application id (this is how e.g. Openrouter can tell you how popular Openclaw is, and which models are used the most).

Anthropic just checks for that.

[filleokus]

Either me or you are misunderstanding the situation. A comment from the GP link: https://news.ycombinator.com/item?id=47633867

> This is slightly different from what OpenCode was banned from doing; they were a separate harness grabbing a user’s Claude Code session and pretending to be Claude Code.

> OpenClaw was still using Claude Code as the harness (via claude -p)[0]. I understand why Anthropic is doing this (and they’ve made it clear that building products around claude -p is disallowed) but I fear Conductor will be next.

[stingraycharles]

If Openclaw was still using Claude Code as the harness, I don't know how to reconcile that with "Openclaw is based on the pi framework", which is decidedly NOT claude code.

From what I understand, they still had the Claude Code harness available, but were mostly fully integrated on the pi agent framework, using Claude Code's oauth credentials directly,

[piazz]

Openclaw allows you to effectively “shell out” to another harness for your model calls, while still using Pi as your main agentic harness. This is the claude -p workflow. Tools and skills are injected into Claude and they hack session persistence into it as well.

They also absolutely blocked OpenClaw system prompts from this path in the prior weeks, based purely on keyword detection. Seems they’ve undone that now.

[throwpoaster]

No, if you ran Openclaw using Anthropic API as a provider, or had it use the ‘claude -p’ cli interface, you got an email from Anthropic threatening a ban unless you upgraded billing.

This was widely reported, and happened to me. You probably can’t reproduce it or see it in docs because they seem to have changed the policy.

[holoduke]

Claude -p is using Claude cli. How can you know it's my own claw?

[ElFitz]

And yet running the Claude Code cli with `-p` in ephemeral VMs gets me the "Third-party apps now draw from extra usage, not plan limits. We've added a credit to your organization to get you started. Ask your workspace admin to claim it and keep going." error.

One day you're experimenting just fine. The next, everything breaks.

And I'd gladly use their web containerized agents instead (it would pretty much be the same thing), but we happen to do Apple stuff. So unless we want to dive into relying on ever-changing unreliable toolchains that break every time Apple farts, we're stuck with macOS.